Due to their random nature, obtaining reliable models that can describe the behaviour of waves is far from simple. This paper presents an approach for forecasting the capabilities of wave energy converters (WECs) for two points, one of them located offshore and the other nearshore. Bivariate Weibull distributions were fitted from spectral significant wave height and mean peak period data. Then, models relating the parameters of these distributions to the day of the year were obtained using mixture density networks, which give the distribution of the predicted variables instead of their expected value. Energy conversion capabilities were forecasted by generating a set of random values for the bivariate Weibull coefficients from the modelled distributions for the period in question. Predicted cumulative distributions for spectral significant wave heights and mean peak periods were then combined with the matrix of the converter in question, allowing the corresponding energy conversion capability to be computed. The proposed method was validated by considering data from the last three years, which were not used to train the models. The resulting predictions were consistent not only with the expected seasonal behaviour, but also with the expected differences between the offshore and nearshore points. It should be also noted that all the validation energy values fall into the forecasted 95% confidence intervals, showing the effectiveness of the approach.